A correlation method based on a correlation calculation and a block matching method are known as conventional means for detecting a motion vector between a plurality of images. In the block matching method, an input image signal is divided into a plurality of appropriately sized blocks (for example, 8 pixels×8 lines), a difference with a pixel located within a fixed range of a previous frame is calculated in units of these blocks, and a block position in which the sum of an absolute value of the difference is at a minimum is found. A relative deviation between images represents the motion vector of the block. A representative inter-frame motion vector is then calculated by integrating the motion vectors of the respective blocks and used to perform final image registration.
A method of determining a sum of squared intensity difference (SSD), which is a sum of squares of a difference between pixel values, or a sum of absolute intensity difference (SAD), which is a sum of absolute values of a difference between pixel values, may be employed during block matching as a method of finding a block having a high correlation. In this method, the correlation (alignment) is determined to be steadily higher as the SSD or SAD decreases. In a reference block region I and a subject block region I′ of a matching operation, when pixel values in pixel positions pεI and qεI′ (where p, q are quantities having two-dimensional values, I, I′ are two-dimensional regions, and pεI indicates that a coordinate p is included in the region I) are set as Lp, Lq, respectively, SSD and SAD are expressed respectively by following Equations (1), (2).
                              SSD          ⁡                      (                          I              ,                              I                ′                                      )                          =                              ∑                                          p                ∈                I                            ,                              q                ∈                                  I                  ′                                                              ⁢                                    (                              Lp                -                Lq                            )                        2                                              (        1        )                                          SAD          ⁡                      (                          I              ,                              I                ′                                      )                          =                              ∑                                          p                ∈                I                            ,                              q                ∈                                  I                  ′                                                              ⁢                                                Lp              -              Lq                                                                      (        2        )            
Alternatively, a method using a normalized cross-correlation (NCC) may be employed. As a zero average correlation, average values Ave (Lp), Ave (Lq) of the pixels pεI, qεI′ included respectively in the reference block region I and the subject block region I′ of the matching operation are calculated, and differences with pixel values included in the respective blocks using following Equations (3), (4) are calculated.
                                                        Lp              ′                        =                                          Lp                -                                  Ave                  ⁡                                      (                    Lp                    )                                                                                                                    1                    n                                    ⁢                                                            ∑                                              p                        ∈                        I                                                              ⁢                                                                  (                                                  Lp                          -                                                      Ave                            ⁡                                                          (                              Lp                              )                                                                                                      )                                            2                                                                                                                              p          ∈          I                                    (        3        )                                                                    Lq              ′                        =                                          Lq                -                                  Ave                  ⁡                                      (                    Lq                    )                                                                                                                    1                    n                                    ⁢                                                            ∑                                              q                        ∈                                                  I                          ′                                                                                      ⁢                                                                  (                                                  Lq                          -                                                      Ave                            ⁡                                                          (                              Lq                              )                                                                                                      )                                            2                                                                                                                              q          ∈                      I            ′                                              (        4        )            
Next, the normalized cross-correlation NCC is calculated using a following Equation (5).NCC=ΣLp′Lq′  (5)
A block having a large normalized cross-correlation NCC is determined to be a block having a high degree of alignment (a high correlation), and therefore a relative deviation between the block I and the block I′ having the highest degree of alignment is determined.
Further, a method of detecting a number of edges in horizontal and vertical directions of a block and weighting the detected edges (JP5-233815A) and a method of performing a reliability determination on the basis of the correlation value itself (JP2005-260481A) are known as methods of evaluating the reliability of a detected motion vector.